Learn.fit vs learn.predict


I want to make sure I understand this correctly.
data = ImageClassifierData.from_paths(PATH, tfms=tfms_from_model(arch, sz))
learn = ConvLearner.pretrained(arch, data, precompute=True)
learn.fit(0.01, 2)

First the model is trained on the training set, the accuracy for each epoch is the comparison of predictions from validation set with the actual validation values.
What is the point of learn.predict on the validation set lesson 1? I know learn.predict is used for a test set but does this mean the predictions used to determine accuracy are not stored in the learn object and have to be made again? Will the predictions have different values than the ones obtained when determining accuracy?

I haven’t checked, but I’d expect so
If you predict on the same set of validation data with the same model you’ll get very near the same result. Without knowing where you are in the notebook, you should also check if subsets of the data are used to speed things up and then the whole dataset used (and things like that)